Self Learning and its Implications for Machine Understanding

Contemporary machine intelligence is far from realizing prominent hallmarks of human understanding and consciousness. The primary shortcoming of current methods can be attributed to the difficulty or implausibility of foreseeing and pre-programming each and every piece of information or knowledge. Emergent intelligence methods based on principles of self learning and self organization have been successful in infusing traits of understanding in machines. This understanding is in contrast to the constrained intelligence permeated on machines by classical approaches of intelligence following supervised knowledge acquisition mechanisms. The primary objective of this paper is to review current work in emergent intelligence methods and discuss means of orchestrating these in to a practical model that resembles the process of human understanding. The paper delineates intricacies of self-learning in humans from both biological and psychological perspectives. Following a discussion of several artificial models of the human mind that have been researched and documented at the conceptual level, we propose a comparatively pragmatic approach based on a novel unsupervised learning algorithm, the GSOM algorithm. This algorithm has been successfully applied to many real world knowledge acquisition and pattern discovery problems. The paper concludes with a further discussion of research developments in emergent systems, which we perceive to be the stepping stones in the search for true machine understanding.